Research Of Face Recognition Based On Modified2DPCA And Neural Network

As an effective characteristics identification technology, face recognition technology has more advantages than fingerprinting. So face recognition technology get a rapid development in recent years.But, the effect of environment has a direct impact on the performance of face recognition systems. For example, age change, low quality photos, etc. In addition, the problems produced in face recognition problem are not alone but many problems combined, for example, obstacles and light problems appear at the same time, which will further increase the difficulty of the face recognition. The work of this paper closely around above problems, and put forward alternative solutions:1) In order to solving main problems of the principal component analysis (PCA) such as large operation data, more memory, the processing time long, the two dimensional principal component analysis (2DPCA) method is proposed in the literature, which doesn’t need to put the picture matrix into Id matrix, but directly solve the covariance matrix. In this thesis, through the matlab simulation platform verification, it is verified that2DPCA has the absolute advantage in the speed of recognition.2) A new improved method is proposed in this thesis which fuses bidirectional2DPCA and PCA methods to further speed up the recognition and increase reconstruction effect. The main difficult point of the improved method is method fusion in the process of simulation.3) A series of experiments with the improved method of PCA and the traditional method of PCA on ORL face database are designed. The test results show that the improved method has clear advantages in processing speed and reconstruction, and the improved method saves a lot of memory.4) This thesis first uses the improved method to process face images, and then fuses the self-learning ability of neural network to deal with face recognition classification. Because of strong self-learning ability and good memory of neural networks, so the new method using neural network is designed to handle large databases. But the design of the neural networks is difficult. If we want to achieve a high recognition rate we have to do a lot of experiments to get a better neural network.Finally, we do the comparison test with those contraction mathods. The experimental comparison results show that the new mathod has a high compression ratio.This thesis takes the British ORL face database as the basis experimental. The database consists of a different time, a black background and details of facial expression changes were composed of400gray images.